Bounded Autonomy on M.F.Bormanhttps://www.mfborman.nl/series/bounded-autonomy/Recent content in Bounded Autonomy on M.F.BormanHugoenThu, 02 Jul 2026 00:00:00 +0000An autonomous agent is safe inside the checks. That is the problemhttps://www.mfborman.nl/posts/2026-07-02-an-autonomous-agent-is-safe-inside-the-checks/Thu, 02 Jul 2026 00:00:00 +0000https://www.mfborman.nl/posts/2026-07-02-an-autonomous-agent-is-safe-inside-the-checks/An autonomous agent stays green while every undeclared obligation it disturbs breaks in silence. Velocity widens the gap, it does not close it. A generator can run the bench but not the floor: discovery, ownership, and who owes what to whom is not a check.An invariant is what others rely on. You author only half of ithttps://www.mfborman.nl/posts/2026-07-02-an-invariant-is-what-others-rely-on/Thu, 02 Jul 2026 00:00:00 +0000https://www.mfborman.nl/posts/2026-07-02-an-invariant-is-what-others-rely-on/An invariant is what others rely on, and you author only half of it. Your consumers mint the rest by depending on you (Hyrum’s Law). A check enforces only the half you declared; the undeclared half is found by exploration, not enforcement.AI moves what you audit, from the decision to the support behind ithttps://www.mfborman.nl/posts/2026-07-02-ai-moves-what-you-audit/Thu, 02 Jul 2026 00:00:00 +0000https://www.mfborman.nl/posts/2026-07-02-ai-moves-what-you-audit/AI cannot take the four judgements behind a boundary. It can assemble the menu of decisions you choose from. So the thing you audit moves up a level: from whether the decision is right to whether the support behind it was complete.Judgement does not end at the choice, it runs the whole linehttps://www.mfborman.nl/posts/2026-07-02-judgement-does-not-end-at-the-choice/Thu, 02 Jul 2026 00:00:00 +0000https://www.mfborman.nl/posts/2026-07-02-judgement-does-not-end-at-the-choice/A boundary splits into four parts, and the last three are not mechanical. Each handoff carries a judgement. So “can AI make the decision” is the wrong question: judgement runs the whole line.The rate at which you turn intent into checked boundaries caps how much you can safely automatehttps://www.mfborman.nl/posts/2026-07-02-the-rate-you-turn-intent-into-checked-boundaries/Thu, 02 Jul 2026 00:00:00 +0000https://www.mfborman.nl/posts/2026-07-02-the-rate-you-turn-intent-into-checked-boundaries/A boundary you can trust is made of four things, and making them explicit is not a documentation task but a change to the work that produces them. The rate at which you turn intent into checked boundaries, not the model or the seat count, caps how much you can safely automate.A boundary an agent cannot see, it walks straight throughhttps://www.mfborman.nl/posts/2026-06-30-a-boundary-an-agent-cannot-see-it-walks-straight-through/Tue, 30 Jun 2026 00:00:00 +0000https://www.mfborman.nl/posts/2026-06-30-a-boundary-an-agent-cannot-see-it-walks-straight-through/Your system already has boundaries. They are just enforced by people remembering them. An agent cannot respect a boundary it cannot see.